SkPath::fFirstDirection: seq-cst -> relaxed

We landed this originally with lazily-correct sequentially-consistent memory
order.  It turns out that's regressed performance, we think particularly when
recording paths.  We also think there's no need for anything but relaxed memory
order here.

We should see this chart go down if all goes well: https://perf.skia.org/#4329
There are also Chrome performance charts to watch in the linked bug.

BUG=chromium:537700

CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-TSAN-Trybot,Test-Ubuntu-GCC-Golo-GPU-GT610-x86_64-Release-TSAN

No public API changes.
TBR=reed@google.com

Review URL: https://codereview.chromium.org/1393833003
diff --git a/include/core/SkPath.h b/include/core/SkPath.h
index d3673b6..2200e03 100644
--- a/include/core/SkPath.h
+++ b/include/core/SkPath.h
@@ -938,13 +938,12 @@
         kCurrent_Version = 1
     };
 
-    SkAutoTUnref<SkPathRef>   fPathRef;
-
-    int                       fLastMoveToIndex;
-    uint8_t                   fFillType;
-    mutable uint8_t           fConvexity;
-    mutable SkAtomic<uint8_t> fFirstDirection;    // SkPathPriv::FirstDirection
-    mutable SkBool8           fIsVolatile;
+    SkAutoTUnref<SkPathRef>                            fPathRef;
+    int                                                fLastMoveToIndex;
+    uint8_t                                            fFillType;
+    mutable uint8_t                                    fConvexity;
+    mutable SkAtomic<uint8_t, sk_memory_order_relaxed> fFirstDirection;// SkPathPriv::FirstDirection
+    mutable SkBool8                                    fIsVolatile;
 
     /** Resets all fields other than fPathRef to their initial 'empty' values.
      *  Assumes the caller has already emptied fPathRef.
diff --git a/include/private/SkAtomics.h b/include/private/SkAtomics.h
index e947d1a..56eace4 100644
--- a/include/private/SkAtomics.h
+++ b/include/private/SkAtomics.h
@@ -42,43 +42,44 @@
 
 // A little wrapper class for small T (think, builtins: int, float, void*) to
 // ensure they're always used atomically.  This is our stand-in for std::atomic<T>.
-template <typename T>
+// !!! Please _really_ know what you're doing if you change default_memory_order. !!!
+template <typename T, sk_memory_order default_memory_order = sk_memory_order_seq_cst>
 class SkAtomic : SkNoncopyable {
 public:
     SkAtomic() {}
     explicit SkAtomic(const T& val) : fVal(val) {}
 
     // It is essential we return by value rather than by const&.  fVal may change at any time.
-    T load(sk_memory_order mo = sk_memory_order_seq_cst) const {
+    T load(sk_memory_order mo = default_memory_order) const {
         return sk_atomic_load(&fVal, mo);
     }
 
-    void store(const T& val, sk_memory_order mo = sk_memory_order_seq_cst) {
+    void store(const T& val, sk_memory_order mo = default_memory_order) {
         sk_atomic_store(&fVal, val, mo);
     }
 
-    // Alias for .load(sk_memory_order_seq_cst).
+    // Alias for .load(default_memory_order).
     operator T() const {
         return this->load();
     }
 
-    // Alias for .store(v, sk_memory_order_seq_cst).
+    // Alias for .store(v, default_memory_order).
     T operator=(const T& v) {
         this->store(v);
         return v;
     }
 
-    T fetch_add(const T& val, sk_memory_order mo = sk_memory_order_seq_cst) {
+    T fetch_add(const T& val, sk_memory_order mo = default_memory_order) {
         return sk_atomic_fetch_add(&fVal, val, mo);
     }
 
-    T fetch_sub(const T& val, sk_memory_order mo = sk_memory_order_seq_cst) {
+    T fetch_sub(const T& val, sk_memory_order mo = default_memory_order) {
         return sk_atomic_fetch_sub(&fVal, val, mo);
     }
 
     bool compare_exchange(T* expected, const T& desired,
-                          sk_memory_order success = sk_memory_order_seq_cst,
-                          sk_memory_order failure = sk_memory_order_seq_cst) {
+                          sk_memory_order success = default_memory_order,
+                          sk_memory_order failure = default_memory_order) {
         return sk_atomic_compare_exchange(&fVal, expected, desired, success, failure);
     }
 private: